Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76033.42691172159,
-      "required_pages_per_second_upper_estimate": 0.005260168895609523
+      "required_token_positions_per_second": 76790.5686850659,
+      "required_pages_per_second_upper_estimate": 0.005312549720300438
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9504.178363965199,
-      "required_pages_per_second_upper_estimate": 0.0039792010111792345
+      "required_token_positions_per_second": 9598.821085633237,
+      "required_pages_per_second_upper_estimate": 0.004018825942376874
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19008.356727930397,
-      "required_pages_per_second_upper_estimate": 0.003958939422351092
+      "required_token_positions_per_second": 19197.642171266474,
+      "required_pages_per_second_upper_estimate": 0.003998362588405173
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28512.535091895596,
-      "required_pages_per_second_upper_estimate": 0.0017092829426388787
+      "required_token_positions_per_second": 28796.463256899708,
+      "required_pages_per_second_upper_estimate": 0.0017263040025976698
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28512.535091895596,
-      "required_pages_per_second_upper_estimate": 0.0022167241900071088
+      "required_token_positions_per_second": 28796.463256899708,
+      "required_pages_per_second_upper_estimate": 0.002238798355968165
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28512.535091895596,
-      "required_pages_per_second_upper_estimate": 0.3657281857838611
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28796.463256899708,
+      "required_pages_per_second_upper_estimate": 0.36937011142622217
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10792501031245649,
+    "producer_batch_seconds": 0.10840794935928955,
     "producer_p95_seconds": 0.21280910899986338,
-    "gpu_update_seconds": 0.9696787714908826,
-    "producer_batches_per_second": 9.265692883464862,
-    "gpu_batches_per_second": 1.0312693537288626,
-    "producer_headroom": 8.984745692250021,
+    "gpu_update_seconds": 0.9601179059159449,
+    "producer_batches_per_second": 9.224415791555689,
+    "gpu_batches_per_second": 1.0415387462709675,
+    "producer_headroom": 8.856526773086422,
     "sustainable": true,
-    "buffer_seconds": 60.120083832434716,
+    "buffer_seconds": 60.48742807270453,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347038.4294757644,
+    "generation_bytes_per_second": 331807.6524095087,
     "upload_bytes_per_second": 5331855.523753104,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.46636832432459,
+    "estimated_full_shards_per_hour": 71.19819812025018,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -130,3 +130,4 @@
 {"step": 570100, "loss": 0.5936891436576843, "lr": 0.0003, "grad_norm": 0.1690868053704629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83445911747, "tokens_per_second": 135515.5591265896, "data_wait_s": 2.7539999791770242e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 570200, "loss": 0.587658166885376, "lr": 0.0003, "grad_norm": 0.1683831684998894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83460408739, "tokens_per_second": 151471.25496444936, "data_wait_s": 1.2197000614833087e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 570300, "loss": 0.6701946258544922, "lr": 0.0003, "grad_norm": 0.2391639645788086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83474990127, "tokens_per_second": 151590.5282242714, "data_wait_s": 1.3880000551580451e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 570400, "loss": 0.5762056708335876, "lr": 0.0003, "grad_norm": 0.14501432976856904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83489649800, "tokens_per_second": 153394.0111943038, "data_wait_s": 8.924000212573446e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d2f4a25996b502ebb1544f80696d71298ed18ac136c823e421d4eef5024ed14
-size 282090
+oid sha256:5de0134794ae9de5e9083485c7960ae9ffaf0b55e22e3914dcf6c171ee4fd390
+size 284208